This research aims to address the execution of repetitive, routine and potentially hazardous tasks by robots operating in crewed low Earth orbit, lunar and Mars-based deployments. Current practices in deploying robotic space systems are limited to manual teleoperation of robots by crew in co-located settings, and the use of carefully hand-crafted structured control sequences from ground control. Both approaches are costly in terms of crew time and effort, and are not scalable for long-term, co-robot deployments. The objective of this work is to improve operational efficiency of robotic systems by lowering deployment time, increasing robustness of routine operations, and increasing support for human astronauts by enabling a robot to leverage the input it obtains from human operators to incrementally increase operational autonomy. Our approach focuses on techniques for identifying what information is needed to improve task performance, decision mechanisms for selecting between crew and ground control interactions, development of interface methodologies for task recovery interactions, and algorithmic methods for improving autonomy based on the acquired instructions.